AI Search Performance Metrics: What to Track and How
Traditional search gave you impressions, clicks, and rank. AI search gives you none of those by default — it just mentions your brand, or doesn't. Here are the four metrics that actually tell you how you're performing, and a practical framework for tracking them over time.
Why measurement is hard
55%
average brand overlap between two identical ChatGPT runs
27%
of brands appear in only one run and never again
16%
of brands are stable across all ten runs of the same prompt
Source: Aiso variance study, 19 consumer prompts × 10 runs each.
The four metrics that matter
Most brands start by asking "does ChatGPT mention us?" That's the right instinct, but a single yes/no answer isn't a metric — it's a snapshot. These four measures give you something you can track over time and act on.
Mention rate
How often your brand appears across a defined set of prompts over a period. This is the primary visibility signal — the equivalent of impressions in traditional search.
Run the same prompt set on multiple occasions and count how many responses include your brand name. Divide by total runs to get a percentage. Track this weekly or monthly against a fixed prompt set so changes are comparable.
Sentiment and context accuracy
Whether the AI mentions your brand positively, neutrally, or negatively — and whether the description matches your actual positioning.
Hallucinations matter here. An AI that mentions your brand in the wrong category, with the wrong price point, or alongside a negative qualifier is a different problem from one that simply omits you. Read the surrounding context, not just whether the name appears.
Prompt coverage
Which prompt types and funnel stages your brand shows up in — and which it doesn't.
A brand might be visible on awareness-stage questions ("what is X?") but absent on decision-stage questions ("which X should I buy?"). Mapping coverage by intent reveals where to focus content.
AI-referred click-through
Sessions that originate from an AI assistant rather than a search engine — identifiable in server logs via the OAI-SearchBot or ClaudeBot user-agent, and in GA4 via session source.
This closes the loop between AI visibility and revenue. See our guide on the ChatGPT funnel for the exact server-log methodology.
How to analyze your AI search visibility
Establish a fixed prompt set
Pick 10 to 20 prompts that represent how real buyers ask about your category. Keep the set stable so you can compare runs over time. Rotating prompts constantly makes trends unreadable.
Run each prompt multiple times
AI answers are variable. Aiso's variance research found that two identical ChatGPT prompts share only 55% of recommended brands on average. A single run is a sample, not a fact — run each prompt at least five times and average the results.
Benchmark against competitors
Measure your mention rate relative to the brands that appear in the same responses. Absolute visibility matters less than relative visibility — if your main competitor appears in 80% of runs and you appear in 20%, that gap is the story.
Connect to business outcomes
Match AI visibility trends to traffic from AI user-agents, and traffic trends to downstream conversions. A rising mention rate that produces no referral traffic signals a citation without a link — a different optimization problem.
Tools for monitoring AI search performance
The AI search visibility tool landscape is still early. A first-party benchmark of five tools found significant differences in methodology, prompt coverage, and cadence — choose based on whether you want real conversation data or simulated prompt runs.
Aiso
AI-specific trackingTracks brand visibility across ChatGPT, Gemini, Claude, and Perplexity using a panel of real, opted-in AI conversations rather than simulated prompts. Surfaces mention rate, context, competitors, and trends.
Profound
AI-specific trackingRuns live queries against major AI platforms and reports citation frequency, context, and share of voice versus named competitors.
Peec AI
AI-specific trackingQuery fan-out tool that expands a seed prompt into a larger set of related queries, then tracks brand appearance across them.
Google Search Console
Traditional (partial signal)Still relevant for AI-assisted search: surfaces the queries where Google's AI Overviews cite your content, and the click-through rates those citations produce.
What happens when you ask an AI chatbot for AI visibility tools
The most striking evidence that dedicated AI search metrics are missing comes from the AI systems themselves. In Aiso's panel of millions of anonymized AI conversations, a user asked an AI assistant whether a specific AI visibility platform was the best tool for getting visible in AI searches. The assistant responded by listing traditional SEO platforms — Semrush, Ahrefs, Moz — and then, when the user clarified they wanted tools specifically for AI response visibility, the assistant acknowledged:
"There isn't yet a dominant, specialized 'AI search visibility' tool like traditional SEO suites… AI models (like ChatGPT) generate responses based on training data or live connections to the web, but they don't crawl webpages in the traditional way search engines like Google do. So direct 'ranking' isn't applicable — visibility depends on how well your content is structured and authoritative in trusted data sources."
AI assistant response, from Aiso's panel of millions of anonymized AI conversations.
That response is both accurate and revealing. Because AI visibility doesn't map to traditional ranking, it also doesn't map to traditional metrics. The tools that tell you your Google rank tell you nothing about whether ChatGPT mentions you when a buyer asks for a recommendation. That gap is what AI search performance metrics are built to close.
Starting points
If you're building an AI search performance measurement practice from scratch:
- Start with 10 prompts covering your main category and competitor comparisons. See our guide on how many prompts to track.
- Run each prompt at least five times before reporting — single runs are not statistically reliable.
- Track mention rate and context accuracy as your primary signals. Add click-through once you have server-log attribution in place.
- Benchmark against two or three direct competitors so you have a relative frame of reference.
- Review how often to run prompts to calibrate your cadence to your brand's variability.
See your brand's AI search performance
Aiso tracks mention rate, context accuracy, and competitor share of voice across ChatGPT, Gemini, Claude, and Perplexity — using real opted-in conversations, not simulated prompts.